xref: /openbmc/linux/net/rxrpc/recvmsg.c (revision 11788d9b)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* RxRPC recvmsg() implementation
3  *
4  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/net.h>
11 #include <linux/skbuff.h>
12 #include <linux/export.h>
13 #include <linux/sched/signal.h>
14 
15 #include <net/sock.h>
16 #include <net/af_rxrpc.h>
17 #include "ar-internal.h"
18 
19 /*
20  * Post a call for attention by the socket or kernel service.  Further
21  * notifications are suppressed by putting recvmsg_link on a dummy queue.
22  */
23 void rxrpc_notify_socket(struct rxrpc_call *call)
24 {
25 	struct rxrpc_sock *rx;
26 	struct sock *sk;
27 
28 	_enter("%d", call->debug_id);
29 
30 	if (!list_empty(&call->recvmsg_link))
31 		return;
32 
33 	rcu_read_lock();
34 
35 	rx = rcu_dereference(call->socket);
36 	sk = &rx->sk;
37 	if (rx && sk->sk_state < RXRPC_CLOSE) {
38 		if (call->notify_rx) {
39 			spin_lock_bh(&call->notify_lock);
40 			call->notify_rx(sk, call, call->user_call_ID);
41 			spin_unlock_bh(&call->notify_lock);
42 		} else {
43 			write_lock_bh(&rx->recvmsg_lock);
44 			if (list_empty(&call->recvmsg_link)) {
45 				rxrpc_get_call(call, rxrpc_call_got);
46 				list_add_tail(&call->recvmsg_link, &rx->recvmsg_q);
47 			}
48 			write_unlock_bh(&rx->recvmsg_lock);
49 
50 			if (!sock_flag(sk, SOCK_DEAD)) {
51 				_debug("call %ps", sk->sk_data_ready);
52 				sk->sk_data_ready(sk);
53 			}
54 		}
55 	}
56 
57 	rcu_read_unlock();
58 	_leave("");
59 }
60 
61 /*
62  * Transition a call to the complete state.
63  */
64 bool __rxrpc_set_call_completion(struct rxrpc_call *call,
65 				 enum rxrpc_call_completion compl,
66 				 u32 abort_code,
67 				 int error)
68 {
69 	if (call->state < RXRPC_CALL_COMPLETE) {
70 		call->abort_code = abort_code;
71 		call->error = error;
72 		call->completion = compl,
73 		call->state = RXRPC_CALL_COMPLETE;
74 		trace_rxrpc_call_complete(call);
75 		wake_up(&call->waitq);
76 		rxrpc_notify_socket(call);
77 		return true;
78 	}
79 	return false;
80 }
81 
82 bool rxrpc_set_call_completion(struct rxrpc_call *call,
83 			       enum rxrpc_call_completion compl,
84 			       u32 abort_code,
85 			       int error)
86 {
87 	bool ret = false;
88 
89 	if (call->state < RXRPC_CALL_COMPLETE) {
90 		write_lock_bh(&call->state_lock);
91 		ret = __rxrpc_set_call_completion(call, compl, abort_code, error);
92 		write_unlock_bh(&call->state_lock);
93 	}
94 	return ret;
95 }
96 
97 /*
98  * Record that a call successfully completed.
99  */
100 bool __rxrpc_call_completed(struct rxrpc_call *call)
101 {
102 	return __rxrpc_set_call_completion(call, RXRPC_CALL_SUCCEEDED, 0, 0);
103 }
104 
105 bool rxrpc_call_completed(struct rxrpc_call *call)
106 {
107 	bool ret = false;
108 
109 	if (call->state < RXRPC_CALL_COMPLETE) {
110 		write_lock_bh(&call->state_lock);
111 		ret = __rxrpc_call_completed(call);
112 		write_unlock_bh(&call->state_lock);
113 	}
114 	return ret;
115 }
116 
117 /*
118  * Record that a call is locally aborted.
119  */
120 bool __rxrpc_abort_call(const char *why, struct rxrpc_call *call,
121 			rxrpc_seq_t seq, u32 abort_code, int error)
122 {
123 	trace_rxrpc_abort(call->debug_id, why, call->cid, call->call_id, seq,
124 			  abort_code, error);
125 	return __rxrpc_set_call_completion(call, RXRPC_CALL_LOCALLY_ABORTED,
126 					   abort_code, error);
127 }
128 
129 bool rxrpc_abort_call(const char *why, struct rxrpc_call *call,
130 		      rxrpc_seq_t seq, u32 abort_code, int error)
131 {
132 	bool ret;
133 
134 	write_lock_bh(&call->state_lock);
135 	ret = __rxrpc_abort_call(why, call, seq, abort_code, error);
136 	write_unlock_bh(&call->state_lock);
137 	return ret;
138 }
139 
140 /*
141  * Pass a call terminating message to userspace.
142  */
143 static int rxrpc_recvmsg_term(struct rxrpc_call *call, struct msghdr *msg)
144 {
145 	u32 tmp = 0;
146 	int ret;
147 
148 	switch (call->completion) {
149 	case RXRPC_CALL_SUCCEEDED:
150 		ret = 0;
151 		if (rxrpc_is_service_call(call))
152 			ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ACK, 0, &tmp);
153 		break;
154 	case RXRPC_CALL_REMOTELY_ABORTED:
155 		tmp = call->abort_code;
156 		ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
157 		break;
158 	case RXRPC_CALL_LOCALLY_ABORTED:
159 		tmp = call->abort_code;
160 		ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
161 		break;
162 	case RXRPC_CALL_NETWORK_ERROR:
163 		tmp = -call->error;
164 		ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NET_ERROR, 4, &tmp);
165 		break;
166 	case RXRPC_CALL_LOCAL_ERROR:
167 		tmp = -call->error;
168 		ret = put_cmsg(msg, SOL_RXRPC, RXRPC_LOCAL_ERROR, 4, &tmp);
169 		break;
170 	default:
171 		pr_err("Invalid terminal call state %u\n", call->state);
172 		BUG();
173 		break;
174 	}
175 
176 	trace_rxrpc_recvmsg(call, rxrpc_recvmsg_terminal, call->rx_hard_ack,
177 			    call->rx_pkt_offset, call->rx_pkt_len, ret);
178 	return ret;
179 }
180 
181 /*
182  * Pass back notification of a new call.  The call is added to the
183  * to-be-accepted list.  This means that the next call to be accepted might not
184  * be the last call seen awaiting acceptance, but unless we leave this on the
185  * front of the queue and block all other messages until someone gives us a
186  * user_ID for it, there's not a lot we can do.
187  */
188 static int rxrpc_recvmsg_new_call(struct rxrpc_sock *rx,
189 				  struct rxrpc_call *call,
190 				  struct msghdr *msg, int flags)
191 {
192 	int tmp = 0, ret;
193 
194 	ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NEW_CALL, 0, &tmp);
195 
196 	if (ret == 0 && !(flags & MSG_PEEK)) {
197 		_debug("to be accepted");
198 		write_lock_bh(&rx->recvmsg_lock);
199 		list_del_init(&call->recvmsg_link);
200 		write_unlock_bh(&rx->recvmsg_lock);
201 
202 		rxrpc_get_call(call, rxrpc_call_got);
203 		write_lock(&rx->call_lock);
204 		list_add_tail(&call->accept_link, &rx->to_be_accepted);
205 		write_unlock(&rx->call_lock);
206 	}
207 
208 	trace_rxrpc_recvmsg(call, rxrpc_recvmsg_to_be_accepted, 1, 0, 0, ret);
209 	return ret;
210 }
211 
212 /*
213  * End the packet reception phase.
214  */
215 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
216 {
217 	_enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
218 
219 	trace_rxrpc_receive(call, rxrpc_receive_end, 0, call->rx_top);
220 	ASSERTCMP(call->rx_hard_ack, ==, call->rx_top);
221 
222 	if (call->state == RXRPC_CALL_CLIENT_RECV_REPLY) {
223 		rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, serial, false, true,
224 				  rxrpc_propose_ack_terminal_ack);
225 		//rxrpc_send_ack_packet(call, false, NULL);
226 	}
227 
228 	write_lock_bh(&call->state_lock);
229 
230 	switch (call->state) {
231 	case RXRPC_CALL_CLIENT_RECV_REPLY:
232 		__rxrpc_call_completed(call);
233 		write_unlock_bh(&call->state_lock);
234 		break;
235 
236 	case RXRPC_CALL_SERVER_RECV_REQUEST:
237 		call->tx_phase = true;
238 		call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
239 		call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
240 		write_unlock_bh(&call->state_lock);
241 		rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial, false, true,
242 				  rxrpc_propose_ack_processing_op);
243 		break;
244 	default:
245 		write_unlock_bh(&call->state_lock);
246 		break;
247 	}
248 }
249 
250 /*
251  * Discard a packet we've used up and advance the Rx window by one.
252  */
253 static void rxrpc_rotate_rx_window(struct rxrpc_call *call)
254 {
255 	struct rxrpc_skb_priv *sp;
256 	struct sk_buff *skb;
257 	rxrpc_serial_t serial;
258 	rxrpc_seq_t hard_ack, top;
259 	bool last = false;
260 	u8 subpacket;
261 	int ix;
262 
263 	_enter("%d", call->debug_id);
264 
265 	hard_ack = call->rx_hard_ack;
266 	top = smp_load_acquire(&call->rx_top);
267 	ASSERT(before(hard_ack, top));
268 
269 	hard_ack++;
270 	ix = hard_ack & RXRPC_RXTX_BUFF_MASK;
271 	skb = call->rxtx_buffer[ix];
272 	rxrpc_see_skb(skb, rxrpc_skb_rotated);
273 	sp = rxrpc_skb(skb);
274 
275 	subpacket = call->rxtx_annotations[ix] & RXRPC_RX_ANNO_SUBPACKET;
276 	serial = sp->hdr.serial + subpacket;
277 
278 	if (subpacket == sp->nr_subpackets - 1 &&
279 	    sp->rx_flags & RXRPC_SKB_INCL_LAST)
280 		last = true;
281 
282 	call->rxtx_buffer[ix] = NULL;
283 	call->rxtx_annotations[ix] = 0;
284 	/* Barrier against rxrpc_input_data(). */
285 	smp_store_release(&call->rx_hard_ack, hard_ack);
286 
287 	rxrpc_free_skb(skb, rxrpc_skb_freed);
288 
289 	trace_rxrpc_receive(call, rxrpc_receive_rotate, serial, hard_ack);
290 	if (last) {
291 		rxrpc_end_rx_phase(call, serial);
292 	} else {
293 		/* Check to see if there's an ACK that needs sending. */
294 		if (after_eq(hard_ack, call->ackr_consumed + 2) ||
295 		    after_eq(top, call->ackr_seen + 2) ||
296 		    (hard_ack == top && after(hard_ack, call->ackr_consumed)))
297 			rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial,
298 					  true, true,
299 					  rxrpc_propose_ack_rotate_rx);
300 		if (call->ackr_reason && call->ackr_reason != RXRPC_ACK_DELAY)
301 			rxrpc_send_ack_packet(call, false, NULL);
302 	}
303 }
304 
305 /*
306  * Decrypt and verify a (sub)packet.  The packet's length may be changed due to
307  * padding, but if this is the case, the packet length will be resident in the
308  * socket buffer.  Note that we can't modify the master skb info as the skb may
309  * be the home to multiple subpackets.
310  */
311 static int rxrpc_verify_packet(struct rxrpc_call *call, struct sk_buff *skb,
312 			       u8 annotation,
313 			       unsigned int offset, unsigned int len)
314 {
315 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
316 	rxrpc_seq_t seq = sp->hdr.seq;
317 	u16 cksum = sp->hdr.cksum;
318 	u8 subpacket = annotation & RXRPC_RX_ANNO_SUBPACKET;
319 
320 	_enter("");
321 
322 	/* For all but the head jumbo subpacket, the security checksum is in a
323 	 * jumbo header immediately prior to the data.
324 	 */
325 	if (subpacket > 0) {
326 		__be16 tmp;
327 		if (skb_copy_bits(skb, offset - 2, &tmp, 2) < 0)
328 			BUG();
329 		cksum = ntohs(tmp);
330 		seq += subpacket;
331 	}
332 
333 	return call->security->verify_packet(call, skb, offset, len,
334 					     seq, cksum);
335 }
336 
337 /*
338  * Locate the data within a packet.  This is complicated by:
339  *
340  * (1) An skb may contain a jumbo packet - so we have to find the appropriate
341  *     subpacket.
342  *
343  * (2) The (sub)packets may be encrypted and, if so, the encrypted portion
344  *     contains an extra header which includes the true length of the data,
345  *     excluding any encrypted padding.
346  */
347 static int rxrpc_locate_data(struct rxrpc_call *call, struct sk_buff *skb,
348 			     u8 *_annotation,
349 			     unsigned int *_offset, unsigned int *_len,
350 			     bool *_last)
351 {
352 	struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
353 	unsigned int offset = sizeof(struct rxrpc_wire_header);
354 	unsigned int len;
355 	bool last = false;
356 	int ret;
357 	u8 annotation = *_annotation;
358 	u8 subpacket = annotation & RXRPC_RX_ANNO_SUBPACKET;
359 
360 	/* Locate the subpacket */
361 	offset += subpacket * RXRPC_JUMBO_SUBPKTLEN;
362 	len = skb->len - offset;
363 	if (subpacket < sp->nr_subpackets - 1)
364 		len = RXRPC_JUMBO_DATALEN;
365 	else if (sp->rx_flags & RXRPC_SKB_INCL_LAST)
366 		last = true;
367 
368 	if (!(annotation & RXRPC_RX_ANNO_VERIFIED)) {
369 		ret = rxrpc_verify_packet(call, skb, annotation, offset, len);
370 		if (ret < 0)
371 			return ret;
372 		*_annotation |= RXRPC_RX_ANNO_VERIFIED;
373 	}
374 
375 	*_offset = offset;
376 	*_len = len;
377 	*_last = last;
378 	call->security->locate_data(call, skb, _offset, _len);
379 	return 0;
380 }
381 
382 /*
383  * Deliver messages to a call.  This keeps processing packets until the buffer
384  * is filled and we find either more DATA (returns 0) or the end of the DATA
385  * (returns 1).  If more packets are required, it returns -EAGAIN.
386  */
387 static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
388 			      struct msghdr *msg, struct iov_iter *iter,
389 			      size_t len, int flags, size_t *_offset)
390 {
391 	struct rxrpc_skb_priv *sp;
392 	struct sk_buff *skb;
393 	rxrpc_serial_t serial;
394 	rxrpc_seq_t hard_ack, top, seq;
395 	size_t remain;
396 	bool rx_pkt_last;
397 	unsigned int rx_pkt_offset, rx_pkt_len;
398 	int ix, copy, ret = -EAGAIN, ret2;
399 
400 	if (test_and_clear_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags) &&
401 	    call->ackr_reason)
402 		rxrpc_send_ack_packet(call, false, NULL);
403 
404 	rx_pkt_offset = call->rx_pkt_offset;
405 	rx_pkt_len = call->rx_pkt_len;
406 	rx_pkt_last = call->rx_pkt_last;
407 
408 	if (call->state >= RXRPC_CALL_SERVER_ACK_REQUEST) {
409 		seq = call->rx_hard_ack;
410 		ret = 1;
411 		goto done;
412 	}
413 
414 	/* Barriers against rxrpc_input_data(). */
415 	hard_ack = call->rx_hard_ack;
416 	seq = hard_ack + 1;
417 
418 	while (top = smp_load_acquire(&call->rx_top),
419 	       before_eq(seq, top)
420 	       ) {
421 		ix = seq & RXRPC_RXTX_BUFF_MASK;
422 		skb = call->rxtx_buffer[ix];
423 		if (!skb) {
424 			trace_rxrpc_recvmsg(call, rxrpc_recvmsg_hole, seq,
425 					    rx_pkt_offset, rx_pkt_len, 0);
426 			break;
427 		}
428 		smp_rmb();
429 		rxrpc_see_skb(skb, rxrpc_skb_seen);
430 		sp = rxrpc_skb(skb);
431 
432 		if (!(flags & MSG_PEEK)) {
433 			serial = sp->hdr.serial;
434 			serial += call->rxtx_annotations[ix] & RXRPC_RX_ANNO_SUBPACKET;
435 			trace_rxrpc_receive(call, rxrpc_receive_front,
436 					    serial, seq);
437 		}
438 
439 		if (msg)
440 			sock_recv_timestamp(msg, sock->sk, skb);
441 
442 		if (rx_pkt_offset == 0) {
443 			ret2 = rxrpc_locate_data(call, skb,
444 						 &call->rxtx_annotations[ix],
445 						 &rx_pkt_offset, &rx_pkt_len,
446 						 &rx_pkt_last);
447 			trace_rxrpc_recvmsg(call, rxrpc_recvmsg_next, seq,
448 					    rx_pkt_offset, rx_pkt_len, ret2);
449 			if (ret2 < 0) {
450 				ret = ret2;
451 				goto out;
452 			}
453 		} else {
454 			trace_rxrpc_recvmsg(call, rxrpc_recvmsg_cont, seq,
455 					    rx_pkt_offset, rx_pkt_len, 0);
456 		}
457 
458 		/* We have to handle short, empty and used-up DATA packets. */
459 		remain = len - *_offset;
460 		copy = rx_pkt_len;
461 		if (copy > remain)
462 			copy = remain;
463 		if (copy > 0) {
464 			ret2 = skb_copy_datagram_iter(skb, rx_pkt_offset, iter,
465 						      copy);
466 			if (ret2 < 0) {
467 				ret = ret2;
468 				goto out;
469 			}
470 
471 			/* handle piecemeal consumption of data packets */
472 			rx_pkt_offset += copy;
473 			rx_pkt_len -= copy;
474 			*_offset += copy;
475 		}
476 
477 		if (rx_pkt_len > 0) {
478 			trace_rxrpc_recvmsg(call, rxrpc_recvmsg_full, seq,
479 					    rx_pkt_offset, rx_pkt_len, 0);
480 			ASSERTCMP(*_offset, ==, len);
481 			ret = 0;
482 			break;
483 		}
484 
485 		/* The whole packet has been transferred. */
486 		if (!(flags & MSG_PEEK))
487 			rxrpc_rotate_rx_window(call);
488 		rx_pkt_offset = 0;
489 		rx_pkt_len = 0;
490 
491 		if (rx_pkt_last) {
492 			ASSERTCMP(seq, ==, READ_ONCE(call->rx_top));
493 			ret = 1;
494 			goto out;
495 		}
496 
497 		seq++;
498 	}
499 
500 out:
501 	if (!(flags & MSG_PEEK)) {
502 		call->rx_pkt_offset = rx_pkt_offset;
503 		call->rx_pkt_len = rx_pkt_len;
504 		call->rx_pkt_last = rx_pkt_last;
505 	}
506 done:
507 	trace_rxrpc_recvmsg(call, rxrpc_recvmsg_data_return, seq,
508 			    rx_pkt_offset, rx_pkt_len, ret);
509 	if (ret == -EAGAIN)
510 		set_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags);
511 	return ret;
512 }
513 
514 /*
515  * Receive a message from an RxRPC socket
516  * - we need to be careful about two or more threads calling recvmsg
517  *   simultaneously
518  */
519 int rxrpc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
520 		  int flags)
521 {
522 	struct rxrpc_call *call;
523 	struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
524 	struct list_head *l;
525 	size_t copied = 0;
526 	long timeo;
527 	int ret;
528 
529 	DEFINE_WAIT(wait);
530 
531 	trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_enter, 0, 0, 0, 0);
532 
533 	if (flags & (MSG_OOB | MSG_TRUNC))
534 		return -EOPNOTSUPP;
535 
536 	timeo = sock_rcvtimeo(&rx->sk, flags & MSG_DONTWAIT);
537 
538 try_again:
539 	lock_sock(&rx->sk);
540 
541 	/* Return immediately if a client socket has no outstanding calls */
542 	if (RB_EMPTY_ROOT(&rx->calls) &&
543 	    list_empty(&rx->recvmsg_q) &&
544 	    rx->sk.sk_state != RXRPC_SERVER_LISTENING) {
545 		release_sock(&rx->sk);
546 		return -EAGAIN;
547 	}
548 
549 	if (list_empty(&rx->recvmsg_q)) {
550 		ret = -EWOULDBLOCK;
551 		if (timeo == 0) {
552 			call = NULL;
553 			goto error_no_call;
554 		}
555 
556 		release_sock(&rx->sk);
557 
558 		/* Wait for something to happen */
559 		prepare_to_wait_exclusive(sk_sleep(&rx->sk), &wait,
560 					  TASK_INTERRUPTIBLE);
561 		ret = sock_error(&rx->sk);
562 		if (ret)
563 			goto wait_error;
564 
565 		if (list_empty(&rx->recvmsg_q)) {
566 			if (signal_pending(current))
567 				goto wait_interrupted;
568 			trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_wait,
569 					    0, 0, 0, 0);
570 			timeo = schedule_timeout(timeo);
571 		}
572 		finish_wait(sk_sleep(&rx->sk), &wait);
573 		goto try_again;
574 	}
575 
576 	/* Find the next call and dequeue it if we're not just peeking.  If we
577 	 * do dequeue it, that comes with a ref that we will need to release.
578 	 */
579 	write_lock_bh(&rx->recvmsg_lock);
580 	l = rx->recvmsg_q.next;
581 	call = list_entry(l, struct rxrpc_call, recvmsg_link);
582 	if (!(flags & MSG_PEEK))
583 		list_del_init(&call->recvmsg_link);
584 	else
585 		rxrpc_get_call(call, rxrpc_call_got);
586 	write_unlock_bh(&rx->recvmsg_lock);
587 
588 	trace_rxrpc_recvmsg(call, rxrpc_recvmsg_dequeue, 0, 0, 0, 0);
589 
590 	/* We're going to drop the socket lock, so we need to lock the call
591 	 * against interference by sendmsg.
592 	 */
593 	if (!mutex_trylock(&call->user_mutex)) {
594 		ret = -EWOULDBLOCK;
595 		if (flags & MSG_DONTWAIT)
596 			goto error_requeue_call;
597 		ret = -ERESTARTSYS;
598 		if (mutex_lock_interruptible(&call->user_mutex) < 0)
599 			goto error_requeue_call;
600 	}
601 
602 	release_sock(&rx->sk);
603 
604 	if (test_bit(RXRPC_CALL_RELEASED, &call->flags))
605 		BUG();
606 
607 	if (test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
608 		if (flags & MSG_CMSG_COMPAT) {
609 			unsigned int id32 = call->user_call_ID;
610 
611 			ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
612 				       sizeof(unsigned int), &id32);
613 		} else {
614 			unsigned long idl = call->user_call_ID;
615 
616 			ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
617 				       sizeof(unsigned long), &idl);
618 		}
619 		if (ret < 0)
620 			goto error_unlock_call;
621 	}
622 
623 	if (msg->msg_name && call->peer) {
624 		struct sockaddr_rxrpc *srx = msg->msg_name;
625 		size_t len = sizeof(call->peer->srx);
626 
627 		memcpy(msg->msg_name, &call->peer->srx, len);
628 		srx->srx_service = call->service_id;
629 		msg->msg_namelen = len;
630 	}
631 
632 	switch (READ_ONCE(call->state)) {
633 	case RXRPC_CALL_SERVER_ACCEPTING:
634 		ret = rxrpc_recvmsg_new_call(rx, call, msg, flags);
635 		break;
636 	case RXRPC_CALL_CLIENT_RECV_REPLY:
637 	case RXRPC_CALL_SERVER_RECV_REQUEST:
638 	case RXRPC_CALL_SERVER_ACK_REQUEST:
639 		ret = rxrpc_recvmsg_data(sock, call, msg, &msg->msg_iter, len,
640 					 flags, &copied);
641 		if (ret == -EAGAIN)
642 			ret = 0;
643 
644 		if (after(call->rx_top, call->rx_hard_ack) &&
645 		    call->rxtx_buffer[(call->rx_hard_ack + 1) & RXRPC_RXTX_BUFF_MASK])
646 			rxrpc_notify_socket(call);
647 		break;
648 	default:
649 		ret = 0;
650 		break;
651 	}
652 
653 	if (ret < 0)
654 		goto error_unlock_call;
655 
656 	if (call->state == RXRPC_CALL_COMPLETE) {
657 		ret = rxrpc_recvmsg_term(call, msg);
658 		if (ret < 0)
659 			goto error_unlock_call;
660 		if (!(flags & MSG_PEEK))
661 			rxrpc_release_call(rx, call);
662 		msg->msg_flags |= MSG_EOR;
663 		ret = 1;
664 	}
665 
666 	if (ret == 0)
667 		msg->msg_flags |= MSG_MORE;
668 	else
669 		msg->msg_flags &= ~MSG_MORE;
670 	ret = copied;
671 
672 error_unlock_call:
673 	mutex_unlock(&call->user_mutex);
674 	rxrpc_put_call(call, rxrpc_call_put);
675 	trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
676 	return ret;
677 
678 error_requeue_call:
679 	if (!(flags & MSG_PEEK)) {
680 		write_lock_bh(&rx->recvmsg_lock);
681 		list_add(&call->recvmsg_link, &rx->recvmsg_q);
682 		write_unlock_bh(&rx->recvmsg_lock);
683 		trace_rxrpc_recvmsg(call, rxrpc_recvmsg_requeue, 0, 0, 0, 0);
684 	} else {
685 		rxrpc_put_call(call, rxrpc_call_put);
686 	}
687 error_no_call:
688 	release_sock(&rx->sk);
689 error_trace:
690 	trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
691 	return ret;
692 
693 wait_interrupted:
694 	ret = sock_intr_errno(timeo);
695 wait_error:
696 	finish_wait(sk_sleep(&rx->sk), &wait);
697 	call = NULL;
698 	goto error_trace;
699 }
700 
701 /**
702  * rxrpc_kernel_recv_data - Allow a kernel service to receive data/info
703  * @sock: The socket that the call exists on
704  * @call: The call to send data through
705  * @iter: The buffer to receive into
706  * @want_more: True if more data is expected to be read
707  * @_abort: Where the abort code is stored if -ECONNABORTED is returned
708  * @_service: Where to store the actual service ID (may be upgraded)
709  *
710  * Allow a kernel service to receive data and pick up information about the
711  * state of a call.  Returns 0 if got what was asked for and there's more
712  * available, 1 if we got what was asked for and we're at the end of the data
713  * and -EAGAIN if we need more data.
714  *
715  * Note that we may return -EAGAIN to drain empty packets at the end of the
716  * data, even if we've already copied over the requested data.
717  *
718  * *_abort should also be initialised to 0.
719  */
720 int rxrpc_kernel_recv_data(struct socket *sock, struct rxrpc_call *call,
721 			   struct iov_iter *iter,
722 			   bool want_more, u32 *_abort, u16 *_service)
723 {
724 	size_t offset = 0;
725 	int ret;
726 
727 	_enter("{%d,%s},%zu,%d",
728 	       call->debug_id, rxrpc_call_states[call->state],
729 	       iov_iter_count(iter), want_more);
730 
731 	ASSERTCMP(call->state, !=, RXRPC_CALL_SERVER_ACCEPTING);
732 
733 	mutex_lock(&call->user_mutex);
734 
735 	switch (READ_ONCE(call->state)) {
736 	case RXRPC_CALL_CLIENT_RECV_REPLY:
737 	case RXRPC_CALL_SERVER_RECV_REQUEST:
738 	case RXRPC_CALL_SERVER_ACK_REQUEST:
739 		ret = rxrpc_recvmsg_data(sock, call, NULL, iter,
740 					 iov_iter_count(iter), 0,
741 					 &offset);
742 		if (ret < 0)
743 			goto out;
744 
745 		/* We can only reach here with a partially full buffer if we
746 		 * have reached the end of the data.  We must otherwise have a
747 		 * full buffer or have been given -EAGAIN.
748 		 */
749 		if (ret == 1) {
750 			if (iov_iter_count(iter) > 0)
751 				goto short_data;
752 			if (!want_more)
753 				goto read_phase_complete;
754 			ret = 0;
755 			goto out;
756 		}
757 
758 		if (!want_more)
759 			goto excess_data;
760 		goto out;
761 
762 	case RXRPC_CALL_COMPLETE:
763 		goto call_complete;
764 
765 	default:
766 		ret = -EINPROGRESS;
767 		goto out;
768 	}
769 
770 read_phase_complete:
771 	ret = 1;
772 out:
773 	switch (call->ackr_reason) {
774 	case RXRPC_ACK_IDLE:
775 		break;
776 	case RXRPC_ACK_DELAY:
777 		if (ret != -EAGAIN)
778 			break;
779 		/* Fall through */
780 	default:
781 		rxrpc_send_ack_packet(call, false, NULL);
782 	}
783 
784 	if (_service)
785 		*_service = call->service_id;
786 	mutex_unlock(&call->user_mutex);
787 	_leave(" = %d [%zu,%d]", ret, iov_iter_count(iter), *_abort);
788 	return ret;
789 
790 short_data:
791 	trace_rxrpc_rx_eproto(call, 0, tracepoint_string("short_data"));
792 	ret = -EBADMSG;
793 	goto out;
794 excess_data:
795 	trace_rxrpc_rx_eproto(call, 0, tracepoint_string("excess_data"));
796 	ret = -EMSGSIZE;
797 	goto out;
798 call_complete:
799 	*_abort = call->abort_code;
800 	ret = call->error;
801 	if (call->completion == RXRPC_CALL_SUCCEEDED) {
802 		ret = 1;
803 		if (iov_iter_count(iter) > 0)
804 			ret = -ECONNRESET;
805 	}
806 	goto out;
807 }
808 EXPORT_SYMBOL(rxrpc_kernel_recv_data);
809 
810 /**
811  * rxrpc_kernel_get_reply_time - Get timestamp on first reply packet
812  * @sock: The socket that the call exists on
813  * @call: The call to query
814  * @_ts: Where to put the timestamp
815  *
816  * Retrieve the timestamp from the first DATA packet of the reply if it is
817  * in the ring.  Returns true if successful, false if not.
818  */
819 bool rxrpc_kernel_get_reply_time(struct socket *sock, struct rxrpc_call *call,
820 				 ktime_t *_ts)
821 {
822 	struct sk_buff *skb;
823 	rxrpc_seq_t hard_ack, top, seq;
824 	bool success = false;
825 
826 	mutex_lock(&call->user_mutex);
827 
828 	if (READ_ONCE(call->state) != RXRPC_CALL_CLIENT_RECV_REPLY)
829 		goto out;
830 
831 	hard_ack = call->rx_hard_ack;
832 	if (hard_ack != 0)
833 		goto out;
834 
835 	seq = hard_ack + 1;
836 	top = smp_load_acquire(&call->rx_top);
837 	if (after(seq, top))
838 		goto out;
839 
840 	skb = call->rxtx_buffer[seq & RXRPC_RXTX_BUFF_MASK];
841 	if (!skb)
842 		goto out;
843 
844 	*_ts = skb_get_ktime(skb);
845 	success = true;
846 
847 out:
848 	mutex_unlock(&call->user_mutex);
849 	return success;
850 }
851 EXPORT_SYMBOL(rxrpc_kernel_get_reply_time);
852